The majority of the participants in the experimental group in the study by Brehmer and Jansson ( 1993) failed to reach an acceptable level of performance, despite the fact that they had been given specific target values for several of the central variables in the MORO system. In the discussion of this result, Jansson ( 1997) assumed that it is not enough to know what subgoals to achieve, but more important, to know how to achieve them. This means that the participants must establish some sort of procedural knowledge in the form of a good enough operative mental model to be able to specify action alternatives on a concrete level. However, an ordinary participant in an experiment with MORO seems unlikely to be able to develop such an operative mental model of the system. Jansson ( 1994a) showed, however, that if participants are taught adaptive strategies for how to approach a system like MORO, they can benefit from such instructions and perform significantly better than a control group of participants. The better performances were accompanied by enhanced mental models of the system within these participants. It thus seems as if the complex and opaque nature of MORO demands from the participant a strategy for dealing with such task characteristics in order to be able to develop a good enough operative mental model.
Without such instructions, however, the majority of participants seem to not develop appropriate decision strategies. An important research question then, is whether the absence of adaptive decision strategies leads to bad performance. The results from Jansson ( 1994b) suggest that. Six out of seven maladaptive decision strategies were found more often among participants who later ended up with bad performance. More important, all seven behaviors were found before the participants had any indication of any negative consequences, in terms of external feedback cues, to which they could relate their own actions.
In order to understand human decision making in complex, opaque, and dynamic environments, the results from the studies reviewed above suggest that it is not only important to analyse the performance of the participants, but also to analyse the participants' behaviors.
Björkman ( 1984) concluded that "man's limited cognition in conjunction with technological means of control sometimes may lead to decisions with disastrous consequences in the future" (p. 45). With that conclusion, he not only predicted a number of incidents in real life (for examples of such man-machine failures, see Reason, 1990), but he also predicted the kind